PointRCNN: 3D Object Proposal Generation and Detection From Point Cloud

Abstract

In this paper, we propose PointRCNN for 3D object detection from raw point cloud. The whole framework is composed of two stages: stage-1 for the bottom-up 3D proposal generation and stage-2 for refining proposals in the canonical coordinates to obtain the final detection results. Instead of generating proposals from RGB image or projecting point cloud to bird's view or voxels as previous methods do, our stage-1 sub-network directly generates a small number of high-quality 3D proposals from point cloud in a bottom-up manner via segmenting the point cloud of the whole scene into foreground points and background. The stage-2 sub-network transforms the pooled points of each proposal to canonical coordinates to learn better local spatial features, which is combined with global semantic features of each point learned in stage-1 for accurate box refinement and confidence prediction. Extensive experiments on the 3D detection benchmark of KITTI dataset show that our proposed architecture outperforms state-of-the-art methods with remarkable margins by using only point cloud as input. The code is available at

A review on object pose recovery: From 3D bounding box detectors to full 6D pose estimators

“…Unlike the previous categories of methods, i.e., classification-based and regressionbased, this category performs the classification and regression tasks within a single architecture. The methods can firstly do the classification, the outcomes of which are cured in a regression-based refinement step [105], [84], [78], [166] or vice versa [75], or can do the classification and regression in a single-shot process [87], [145], [101], [106], [100], [148], [103], [102], [30], [37], [162].…”

SARPNET: Shape attention regional proposal network for liDAR-based 3D object detection

“…There have been many two-stage works [31,26,32,22,2] recently. Fast Point R-CNN [2] applies two-stage framework exploiting volumetric representation for initial predictions and raw point cloud for refinement.…”

“…Methods should be at least 20Hz since onboard application should cover 360 degree rather than KITTI annotation at limited 90 degree. Drawn methods are FP: F-PointNet [20], AF: AVOD-FPN [9], M: MMF [13], I: IPOD [31], FC: F-ConvNet [26], S: STD [32], PR: PointRCNN [22], FPR: Fast Point R-CNN [2], SE: SECOND [28], PP: PointPillars [10], PI: PIXOR++ [29] and O: our HVNet. For PointPillars we use their runtime on PyTorch for a fair comparison.…”

MeteorNet: Deep Learning on Dynamic 3D Point Cloud Sequences

“…Point clouds do not suffer as much from quantization errors compared to other geometric representations such as grids. Recently, deep architectures have been proposed that directly consume a single or a pair of point clouds for various 3D recognition tasks [21,22,20,14,24]. These architectures have outperformed methods based on other geometric representations.…”